Like many transmembrane proteins, determination of the structure of P-gp by X-ray crystallography has proven elusive despite much effort. This stems from the difficulty of forming sufficient-quality crystals that also maintain the native physiochemical environments for the different parts of the protein. Thus, in lieu of the direct, experimental determination, we strive to integrate all available (often indirect) experimental data with physiochemically-based mathematical methods to produce one or more physically realistic models of the structure. Fortunately, over three decades of study has provided a wealth of information about P-gp from which we can gleam structural information. On a broad scale, it is known that the protein is composed of two homologous domains, each with a six-segment transmembrane component and a nucleotide-binding component. To date, the best two sources of structural information about P-gp are an X-ray crystal structure of the homologous bacterial protein Sav1866, and low-resolution cryo-electron micrographs of human P-gp. In addition, the corrected X-ray structure of the bacterial lipid flippase MsbA is expected soon, which is even closely related to P-gp than is Sav1866. Thus, one major focus of our work is to develop a homology model of human P-gp using the crystal structures of the bacterial proteins as templates. While such a model has recently been published by Peter Tielemans group, it was only based on a simple alignment of the P-gp and Sav1866 sequences, and unfortunately, this is overall not well defined for the transmembrane segments. Rather, our efforts go deeper into examining the patterns of residue conservation within the family of closely related MDR proteins and the superfamily of ABC transporters. This information helps predict which residues are exposed to the core and headgroup layers of the membrane, which residues line the pore, and which are at the interfaces of the two transmembrane domains. We are currently in the process of developing a grand sequence alignment of homologous families and the superfamily. The results of this will also enable the determination of patterns of correlated mutations, which help identify groups of residues that are proximal in the 3-dimensional structure of the protein. Finally, we will examine the resultant model for consistency with all the experimental data, such as the effects of site-directed mutagenesis, naturally occurring polymorphisms, and cross-linking data. Where the model based on the Sav1866 template fails to explain the experimental results, we will search for alternate conformations that bring it into compliance. The other major focus is to develop models from the density maps obtained from electron microscopy. This has the advantage that the structural data is directly from human P-gp, the target protein, and not from a bacterial homolog, which likely differs in structure to some degree. This includes the fact that the two transmembrane domains of human P-gp are different in sequence, and thus are asymmetrical around the approximate two-fold axis of the pore, while the bacterial homologs only contain one domain, and thus form perfectly symmetrical homodimers in the membrane. We have already contacted the group that published the microscopy data, and have obtained their coordinates for standard protein helices and nucleotide-binding domains fitted to the electron density. However, due to the low resolution of the data, these models only specify the peptide backbone, and not the type of the residue at each position. To solve this, we are embarking on a threading project, which predicts the correct alignment of the amino acid sequence on the structure of the backbone. To this end, we are currently adapting our previously developed threading software to the P-gp protein. Each enumerated alignment of sequence to structure will be scored and ranked according to a number of criteria. As described above, this will again measure compliance with experimental data, correct physiochemical environments for the different types of residues, and whether conserved residues and predicted clusters are proximal in space. In addition, this analysis will include the calculated energy from scales of residue contact potentials specifically derived for membrane proteins